Exploring the Effectiveness of Regularized Representation-Based Feature Selection for PCR in Different Years with THz-TDS

Read full text Summarise with VIRAT
Authors: Xiaowen Zha, Chengyong Zheng, Zihao Liu, Qian Li, Zhijing Ye
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Pericarpium Citri Reticulatae (PCR) of longer storage years generally possesses higher medicinal value. However, it often challenging to distinguish them through traditional sensory analysis methods. In recent years, Terahertz Time-Domain Spectroscopy (THz-TDS) technology has been widely applied in food and drug identification, and feature selection methods based on regularized representation have also been extensively investigated. Therefore, this paper probes the feasibility of feature selection methods based on regularized representation for identifying PCR in differentyears by using THz-TDS. The average accuracy and standard deviation are adopted to assess the performance of the feature selection methods. The main exploration process is as follows: Firstly, we make use of the absorption coefficient and refractive index spectra with the frequency ranges of 0.1-1.0 THz and 0.1-1.5 THz respectively. Secondly, the Support Vector Machine (SVM) is employed as the classifier to establish classification methods in combination with feature selection methods based on regularized representations, which are then applied to identify PCR in different years. Furthermore, we propose a feature selection method based on sparse Low-Rank and Graph Structure Learning (LRGSL), which combines the advantages of feature selection methods based on regularized representation in recent years. The experimental results indicate that the feature selection method based on regularization representation is conducive to enhancing the accuracy of identifying the storage years of PCR, further demonstrating the feasibility of THz-TDS technology combined with the feature selection method based on regularization representation. Moreover, LRGSL, Joint Low-Rank Decomposition and Local Preservation (JLRDLP), and Low-Rank approximation and Structural Learning (LRSL) methods exhibit superior performance in the discrimination of PCR storage years.

Fields of Study & Subjects

Terahertz technology and applications Spectroscopy and Chemometric Analyses Traditional Chinese Medicine Analysis

Publication details

Journal / source
Contemporary Mathematics
Pages
2332–2350
Publication date
1 April 2026
Language
EN